../

Meeting with Kevin

CVA6 mining

Mining different rules Where did these templates come from?

  • LLM based on bug reports and general form
  • T3 not producing rules?
  • T3 is mined used Texada
    • Use the sped-up form
  • The templates don’t match the description but they may be useful in general as they are mining rules

Commit stage has an awful lot of rules for some templates. Can grouping mechanism help here? Grouping is just done with signals now. No static analysis. Implement this in Hardware using counters. This seems scalable.

svabench

  • Any way to incorporate skills?
  • Incorporate documents in the mix
  • Deeper assertions?
    • Have a top model parse the document and instruct the smaller model to extract the same from RTL
  • Test out the correctness